PRELIMINARY EVALUATION OF A COMMERCIAL 360 MULTI-CAMERA … · PRELIMINARY EVALUATION OF A...

__________________________

* Corresponding author

PRELIMINARY EVALUATION OF A COMMERCIAL 360 MULTI-CAMERA RIG FOR

PHOTOGRAMMETRIC PURPOSES

L. Teppati Losè 1*, F. Chiabrando1, A. Spanò 1 1 Politecnico di Torino, Dept. of Architecture and Design – Geomatics for Cultural Heritage Laboratory (LabG4CH)

(lorenzo.teppati, filiberto.chiabrando, antonia.spano)@polito.it

Commission II, WG II/8

KEY WORDS: Calibration, 360 Images, Multi-Camera system, Cultural Heritage, Photogrammetry

ABSTRACT:

The research presented in this paper is focused on a preliminary evaluation of a 360 multi-camera rig: the possibilities to use the

images acquired by the system in a photogrammetric workflow and for the creation of spherical images are investigated and different

tests and analyses are reported. Particular attention is dedicated to different operative approaches for the estimation of the interior

orientation parameters of the cameras, both from an operative and theoretical point of view. The consistency of the six cameras that

compose the 360 system was in depth analysed adopting a self-calibration approach in a commercial photogrammetric software

solution. A 3D calibration field was projected and created, and several topographic measurements were performed in order to have a

set of control points to enhance and control the photogrammetric process. The influence of the interior parameters of the six

cameras were analyse both in the different phases of the photogrammetric workflow (reprojection errors on the single tie point, dense

cloud generation, geometrical description of the surveyed object, etc.), both in the stitching of the different images into a single

spherical panorama (some consideration on the influence of the camera parameters on the overall quality of the spherical image are

reported also in these section).

1. INTRODUCTION

In recent years, the commercial world of entertainment (that is

constantly and rapidly changing) has gone through some new

major transformations, both from the customers side and from

the major companies involved in the market. More specifically,

the biggest changes can be identified in the segment related to

the production and sharing of multimedia immersive contents,

e.g. the 360 images and videos. Researchers in the field of

Geomatics have monitored the evolution of the market, in order

to stress the possibility of using these new systems for metric

documentation purposes, with the typical adopted

methodologies of their discipline.

1.1 Rise of 360 multi-cameras systems

Different technological observers have defined 2017 as the year

of 360 cameras and for several reasons this affirmation can be

considered true: the past year brought important updates, the

launch of several new cameras with different range of prices and

some major technological improvements. Consequently, related

to the new diffusion of sensors, the development or

implementation of new platforms for the use and diffusion of this

contents is growing as well. However, as is well known, this is

not a brand-new technique: the idea of combining together

different images to create wide format omni-comprehensive

representation of the physical space dates back to the late XIX

century and has constantly evolved during time. Also, the use of

these panoramic images for photogrammetric purposes has a

history that dates back in time (Luhmann, 2004) .

Single and multi-camera systems devoted to the immersive

recording of the physical space were already developed and

used before this recent technological acceleration, also thanks to

the implementation of new algorithms for the image stitching

and the massive use of digital technologies (lot of these

improvement can be derived from the computer disciplines as

reporte in Szeliski, 2006). The major and most interesting last

advancements in this technique are related with the

development of ad hoc systems and software with reduced cost

and with the possibility to grant a wider range of people the

accessibility to this kind of technology.

1.2 Action and 360 cameras in photogrammetry

Among the different COTS (Commercial Off The Shelf) sensors

available two categories have been particularly investigated by

Geomatics researchers in the last years: action cameras and 360

cameras. The use of action cameras for photogrammetric

purposes have been investigated by a large number of researchers

in the last years (Balletti, Guerra, Tsioukas, & Vernier, 2014;

Barazzetti, Previtali, & Roncoroni, 2017b; Markiewicz, Łapi,

Bienkowski, & Kaliszewska, 2017; Perfetti, Polari, & Fassi,

2017) and several issues related with these sensors have been

considered: fisheye lenses distortion, short focal lens influence,

mathematical model adopted for the calibration, etc.

On the other hand, the research on 360 cameras has grown in

interest recently, due to the release on the market of new and low-

cost sensors. More complex and expensive systems, e.g. Ladybug

by FLIR, the 360 camera of Google by Immersive Media, iSTAR

by NCTech or images composed by DSRL (Digital single lens

reflex) wide angle cameras, already existed and have been used

but the availability of new less expensive platforms, gave new

launch to the research in this sector. In this case, the attention of

the researchers (Barazzetti, Previtali, & Roncoroni, 2017a; Fangi,

2015; Fangi & Nardinocchi, 2013; Holdener, Nebiker, & Blaser,

2017; Kossieris, Kourounioti, Agrafiotis, & Georgopoulos, 2017)

was focused on the issues related with the system configuration,

the stitching of images, the different approaches to use compared

with traditional photogrammetry, etc.

1.3 The tested system

In the research presented in this contribute a commercial system,

that combine the research on action and 360 camera, was tested: the

Freedom 360 (www. freedom360.us). The system (Figure 1) is

composed by a 3D printed mount design to hold six action cameras

(GoPro Hero4 or Hero3) coupled in opposite positions to record full

spherical immersive videos or images. The main attractive features

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2, 2018 ISPRS TC II Mid-term Symposium “Towards Photogrammetry 2020”, 4–7 June 2018, Riva del Garda, Italy

This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-2-1113-2018 | © Authors 2018. CC BY 4.0 License.

1113

of this system are related to different characteristics: the relative low

cost of the Freedom 360 compared with other similar systems, its

portability (size less than 10x10x10cm and total weight with

cameras of 525 g), the possibility of controlling individually and

independently the six cameras and consequently to have the chance

to manage and process the data recorded from the six cameras both

separately or together.

Figure 1. The Freedom 360 system.

There are obviously some drawbacks: the acquisition and processing

of the data collected by this system of sensors is less controlled,

compared to other more expensive solutions, and is not always easy

to reach good results during the creation of 360 contents. Moreover,

the overall quality of the system is directly influenced by the

characteristics of the single camera. Due to the reasons mentioned

above and to the intrinsic characteristics of the system one of the main

aim of the presented research is to assess and verify the consistency of

the six action cameras, and consequently of the system itself, before

performing some tests in the contest of the documentation of a real

Built Heritage environment.

2. THE PROPOSED METHODOLOGY

The first tests were carried out in order to verify the consistency

of the six cameras: as is well documented by other works

(Balletti et al., 2014; Barazzetti et al., 2017b; Läbe & Förstner,

2004), low cost commercial cameras are equipped with less

stable lenses and the camera parameters need to be carefully

considered. These issues need to be in depth investigated and

can have an impact on different aspects related with the use of

this kind of system. The parameters of the camera’s sensors can

be useful in the generation of good quality panoramic images,

derived from the stitching of different acquisitions together and,

as is well known (Fraser, 2013; Luhmann, Fraser, & Maas,

2016; Schneider, Schwalbe, & Maas, 2009), in case of

photogrammetric approaches they are crucial (in order to

achieve more reliable and metric controlled results).

2.1 Employed system and 3D calibration environment

For our test the Freedom 360 was equipped with six GoPro

Hero 4 Silver Edition, main specifications of the camera are

reported in Table 1. GoPro Hero 4 Silver main specifications

Weight 84 gr

Size 54x41x30 mm

Sensor CMOS – 12MP

Sensor size 1/2.3"

Focal length 2.92 mm

Image resolution Max 4000x3000

Table 1. GoPro Hero 4 Silver main specifications.

To estimate and evaluate these factors different tests for the

calibration of the cameras were carried out. Then, a specific 3D

calibration field was projected and realised (Figure 2, a). The

3D field was created taking into consideration different factors:

presenting a marked 3D component with objects at different

depth, having recognizable features to be used in the

photogrammetric process, simulate both an indoor/outdoor

environment with a short (and as much constant as possible)

acquisition distance. On the calibration field a set of targets was

homogenously distributed in order to use the 3D coordinates of

that measured points for the calibration process. The 14 targets

were measured using a typical traditional approach: two points

forward intersection in order to obtain with a good accuracy the

coordinates of the points. For the topographic survey a Leica

Viva TS16 Total Station was used: accuracy of 1” (0.3 mgon)

on angular measurement and distance accuracy on prism of

1mm+1.5 parts per million. In order to obtain the best accuracy

and precision as possible in the measurement of the distance a

circular mini-prism were used during the topographic survey.

The data collected on the field were then adjusted using

MicroSurvey STAR*NET software (Figure 2, b) where the

planimetric and altimetric components of the forward

intersection were separately considered. According to the

acquisition geometry, distances and adopted strategy the

residual on the 14 targets for both the components planimetric

and altimetric is less than 2 mm. The final 3D coordinates were

then used in the photogrammetric approach both as GCPs

(Ground Control Points) and CPs (Check Points) to precisely

estimate and control the camera interior parameters and to scale

and georeference the generated models.

(a) (b)

Figure 2. The calibration field (a); top view of the compensated

topographic network (b).

2.2 Proposed approach and acquisitions

In order to verify the consistency of the system and estimate the

interior orientation parameters of each one of the cameras, the

six GoPro were marked with an identification letter and

different photogrammetric acquisition were performed:

- Each one of the six cameras was detached from the

rig and inserted on an ad hoc realized 3D printed

support in order to mount it on a photographic tripod

for image acquisition. With this configuration, a

dense set of images was acquired for each camera

(with different camera orientation and relative

position). An average number of 150 images were

obtained for each of the six cameras.

- All the six cameras were mounted on the Freedom

360 rig and used in the time-controlled modality for

the shoot. The system was then moved in different

preselected positions to acquire the whole calibration

field. The images were then selected and only the

images acquired in the chosen positions were

considered. A total of 258 images were acquired with

the 360 configuration, corresponding to 43 pre-

selected position of the rig.

An example of the two networks of images acquired is reported

below in Figure 3.



1114

(a) (b)

Figure 3. Part of the images network acquired on the calibration

field. One of the six camera (a) and the six cameras mounted on

the 360 rig (b).

2.3 Interior orientation parameters estimation

In the presented research a self-calibration approach (Fraser,

1997; Gruen & Beyer, 2001; Remondino & Fraser, 2006) was

adopted in order to estimate the interior orientation parameters

of every camera and analyse their consistency, using a well-

known commercial software solution for SfM (Structure from

Motion): Agisoft Photoscan.

The above-mentioned acquisitions were processed in eight

different projects, divided as following:

- Six individual projects (one for each action camera

mounted on the photographic tripod).

- One project for the images acquired with the 360 rig,

using the native Exif (Exchangeable image file

format) metadata embedded in the cameras.

- One last project for the images acquired with the 360

rig, but applying a modification on the Exif metadata,

in order to have the software recognizing the six

cameras as different.

For each project 8 of the measured target were used as GCPs

while 6 were used as CPs, the RMSe (Root Mean Square error)

on these points is reported in the following Figure 4.

Figure 4. RMSe on GCPs and CPs of the different

photogrammetric projects.

As will be discussed in paragraph 2.3.2, adopting the automated

solution, i.e. using a single calibration model for the six

cameras, will produce poor results in terms of metric accuracy

as well as in the overall quality of the generated 3D model.

Furthermore, all the projects were processed using the same

parameters for the estimation of interior and exterior

orientation, tie point extraction and BBA (Bundle Block

Adjustment): the accuracy of the alignment was set as high in

order to estimate at least three coefficients for the radial

distortion and two for the tangential, plus the focal length, the

principal point coordinates and the skew transformation

coefficients. Key/tie points limit was set at 0 in order to extract

as many points as possible.

2.3.1 Interior orientation evaluation using Six cameras

separately considered

The six action cameras of the 360 system were previously marked

with letters (A, B, C, D, E, F) to create an unique identification of

each GoPro and of their reciprocal positions on the rig. The GCPs

and CPs were placed in all the six projects and the first steps of

the photogrammetric process were completed. After the

estimation of the interior orientation parameters for each camera,

the data were collected and organize in order to be compared and

analysed. The estimated parameters of the six cameras are

reported in Table 2 (Appendix): focal length in pixels and

millimetres (f), principal point coordinates in pixels (cx and cy),

radial distortion coefficients in millimetres (k1, k2, k3), skew

coefficients in pixels (b1 and b2) and tangential distortion

coefficients in millimetres (p1 and p2).

Especially in the case of these low-cost mass market sensors these

parameters need to be carefully considered.

In this case the estimated focal length of the six cameras can be

considered consistent and similar, while some issues can be traced

in the estimation of the principal point coordinates of the different

sensors, i.e. coordinates of lens optical axis interception with

sensor plane (expressed in pixels with cx and cy coefficients).

The estimated principal points of the six cameras sensors is

graphically represented in the following Figure 5.

Figure 5. Estimated principal point coordinates of the six

cameras separately considered and processed.

As is possible to notice, the principal point coordinates of five

out of six cameras can be considered comparable (the deviation

from the ideal principal point of coordinates 0,0 has the same

order of magnitude but located on different quarter of the

sensor) while the camera D presents a complete different

position, almost in the ideal intersection of the two principal

axes. This camera, which value is theoretically closer to the

ideal principal point, can create some issues when working

together with the other five sensors, due to its different interior

parameters. As we will see, this can be considered an issue both

when working with the stitching of the six images in a single

spherical image, both in the photogrammetric process.

2.3.2 Six cameras mounted on the rig and automatically

processed

The images acquired by the six cameras mounted on the rig were

processed both following the automatic workflow implemented in

Photoscan, and in a second time with a manual editing to better

control the camera parameters (paragraph 2.3.3). During the

0,000

0,005

0,010

0,015

0,020

GCPs error (m) CPs error (m)

A (-59,928;-23,881)

B (-76,449;-30,115)C ( 21,457;-44,285)

D ( -1,514; -6,173)

E (48,974; 54,031)

F ( 50,579;22,681)

-80

-60

-40

-20

0

20

40

60

80

-100 -80 -60 -40 -20 0 20 40 60 80 100

Estimated Principal Point (px)

A B C D E F



1115

automatic workflow, the software uses the information derived

from the Exif metadata of the six cameras as initial internal

orientation for the further camera parameters estimation. In this

case, due to the information embedded in the Exif, Photoscan

assumes that only a camera was used and perform all the phases

of the photogrammetric process considering all the cameras as

identical. This wrong assumption lead to different problems in the

phases of interior orientation parameters estimation and tie point

extraction. These issues are evident for example in the value

computed for GCPs and CPs errors (Figure 4). Also, in this case

is interesting to analyse the values estimated for the principal

point coordinates (Figure 6) that are similar to the ones of camera

D, presented in the previous paragraph. The values of the

deviation of the six different cameras from the principal point

coordinates that were evident from the previous estimations are

not considered in this case.


cameras mounted on the rig and automatically processed with

the native Exif information.

The issues derived from this approach are also clearly visible in

another representation: the plotting of the image residuals

reported below in Figure 7.

Figure 7. Image residuals for the 360 configuration

automatically processed with the native Exif.

In this case is clearly shown that the interior parameters

estimated for the camera are not generating consistent results

and that performing a single calibration for the six cameras,

following an automatic approach, is not a successful and

satisfying solution. The other estimated parameters of the

interior orientation are reported in Table 3.

2.3.3 Interior orientation evaluation using Six cameras

mounted on the rig and processed after Exif modification

To evaluate if it was possible to contemporary calibrate, and

with satisfying result, the six cameras mounted on the rig

another project was created and processed. Before importing the

images in the software, the information embedded in the Exif

files were modified and the name of the camera model was

changed, in order to process independently the six cameras

during the workflow. The estimated interior parameters are

reported in Table 4, while the estimated coordinates of the six

principal points are shown in the images below, Figure 8.


cameras mounted on the rig and processed after the Exif

modification.

As shown in Figure 4, adopting this approach is possible to

obtain an RMSe on the GCPs and CPs that can be compared

with the one achieved while working with the six cameras

independently.

3. ANALISYS OF THE RESULTS

The data obtained by the different tested strategies were

analysed and compared, in order to evaluate the impact of the

different possible approaches on the orientation of the cameras

and on the quality of the different generated 3D model. The

consistency of the six low-cost action cameras, the impact of a

correct estimation of the interior parameters, the quality

achievable in the process of tie point extraction and the metric

quality of the 3D models were all considered issues and some

first consideration will be reported in the following paragraphs.

3.1 Cameras consistency

As is well known low-cost cameras, and sensors in general, can

present deformation derived from their mass market production.

The six action cameras considered in this research have

nominally the same exact specifications. Thus, as demonstrated

by the different test performed, each camera has different

characteristic and, in particular, one of the cameras presented a

set of interior parameters that can lead to an inconsistency of the

360 system considered as a whole. This issue can be negligible

if the camera is used as a standalone, as happen in most of the

cases, but can be have a negative impact if the sensor is jointly

used with the other ones. The different tests performed

demonstrated once again the importance of not considering the

photogrammetric software as a black box in which operate fully

automatic procedures. Also in the most diffused commercial

solution, such as Agisoft Photoscan, is possible for the operator

to maintain the control of several parameters during the

different steps of processing. In particular, in the case of low-

cost cameras, and especially if multiple cameras are used

together, is crucial for the operator to adopt some best practices.

Firstly, during the acquisition phases is important to project and

realise a strong network of cameras with a good overlap

between them, to facilitate the phases of interior and exterior

orientation. A simple intervention of the operator, like the

modification of the Exif files, can have a strong impact on the

quality of camera calibration and tie point extraction phases that

influences also the quality of the generated 3D model. The use

of GCPs and CPs is again really important, not only for scaling

(0,381; 1,070)

-80

20

-100 -80 -60 -40 -20 0 20 40 60 80 100

Rig_Native EXIF

A (-59,740; -24,235)

B ( -76,220; -30,213)

C ( 21,893; -44,387)

D ( -1,884; -6,310)

E (48,956; 54,120)

F (50,620; 22,333)

-80

-60

-40

-20

0

20

40

60

80

-100 -80 -60 -40 -20 0 20 40 60 80 100

A B C D E F



1116

and georeferencing the model, but also to aid and optimize

interior and exterior orientation steps.

Comparing the parameters estimated following the three different

approaches (Table 2, Table 3, Table 4), some consideration can

be reported. While is clear that following a fully automatic

procedure lead to poor results, is important to analyse more in

depth how some small interventions can radically change the

generated output. Considering the six separate projects as the

most reliable, is interesting to notice that the parameters reported

in Table 4 can be definitely defined as accurate. The values

estimated in the last of the eight projects for the six cameras

demonstrate that adopting some best practices on the field and in

the post processing phases, allow to calibrate the sensors of the

system in a single photogrammetric project.

3.2 Impact of calibration on image stitching

In the next sections the importance of a good estimation of

camera parameters during the photogrammetric process will be

underlined, but the computed parameters can have an impact

also in other processes related with the 360 cameras output.

One of the main products derived from these systems is the

creation of 360 images or videos, usually represented as an

equirectangular projection derived from the stitching of images

from multiple source. Nowadays, different software solution

exists on the market for the processing of these digital contents,

both opensource (Ptgui, Hugin), and commercial (AutoPano

Giga, VideoStitch). These software presents common features

with the methodological framework of photogrammetry, thus

can be subjected to similar issues connected with the sensors

specifications. The commercial solution tested in this research is

AutoPano Giga (v. 4.2), by Kolor. Using this platform is

possible to create spherical images and it works in subsequent

step: first of all, SIFT (Scale Invariant Feature Transform) or

similar algorithms are used in order to extract tie points from

the used images and an RMSe (Root Mean Square error)

evaluation is also provided for each calculated tie point. During

this process some of the camera parameters are taken into

consideration from the software (i.e. focal length, k1, k2 and k3

coefficient for radial distortion and the coordinates of the

principal point); they are partially read from the Exif file and

partially extracted from the software database. During this part

of the process is possible with some manual editing to modify

the information embedded in the Exif file to let the software

consider separately the six cameras. In this part of the Autopano

workflow the camera calibration parameters extracted in the

self-calibration performed in Photoscan were used. Furthermore

according to the processing steps the stiching process was

realized and the impact on the quality of the process has been

evaluate. In Figure 9 an example of some image aberration that

were corrected thanks to the lens distortion parameters

calculated and imputed for each camera.

Figure 9. Stitching aberration related with the use of different

approach for camera interior parameters estimation.

In Figure 13 the parameters used in the stitching software are

reported: as in Photoscan using the native Exif the software

assumes the same parameters for each camera. Applying again a

modification on the Exif is possible to work with separate

camera characteristic, but Autopano is not able to estimate the

different coefficient with a good level of accuracy. For

improving the stitching quality is possible, as is reported before,

to use the parameters estimated by the photogrammetric

process. There are still some minor issues to fix in the overall

stitching of the spherical images, but, as is possible to notice in

Figure 9, some major image aberrations can be improved

applying the described procedure.

3.3 Impact of calibration on tie points extraction and 3D

model generation

In order to understand and evaluate the impact of a rigorous and

correct calibration of the 360 system on the photogrammetric

process, two other analyses were performed. The first analysis

is related with an assessment of the quality of the extracted tie

points on the two 360 configurations tested: the one

automatically processed with the native Exif file and the one

processed after the Exif modification. Through a Python script

launched by command line it was possible to extract directly

from the photogrammetric project generated in Photoscan a .txt

file formed by four columns and containing some precious

information related with the sparse cloud of tie points. This file

contains for each tie point the spatial coordinates and the

reprojection error. The points were filtered, excluding the so-

called outliners, and all the points with a reprojection error

higher than 10 pixels were not considered. The script was

applied to the two photogrammetric projects and the obtained

data were imported and classified in CloudCompare software,

applying a scale of false colours based on the reprojection error

of each tie point. An extract of these analyses is reported in

Figure 10, where is clearly visible the impact of a correct

calibration on the quality of the tie points extracted in the

photogrammetric process.

Figure 10. Sparse cloud. Tie point quality based on reprojection

error. 360 configuration: automatic process with native Exif

(A), process with modified Exif (B).



1117

In the A configuration is possible to notice how only the 24% of

points presents a reprojection error value minor of 0.5 pixels, while

for the B configuration the 72% of points are comprehend in the same

range of values. If we move the observation to the value of 1 pixel of

reprojection error the ratio between A and B is 40% to 90%.

3.3.1 3D models validation

The influence of a good camera calibration on the generation of

the 3D model is evident also from a first visual quality

assessment between the products derived from the two

configurations (Figure 11). To better understand and analyse the

overall quality of the generated 3D models some more specific

analyses were achieved, using a LiDAR (Light Detection and

Ranging) acquisition as ground truth. The laser dataset was

acquired using a Faro Focus X120 by CAM2, following the

consolidated workflow for acquisition ad post-processing

phases, in particular the scans were registered using a cloud to

cloud approach and then georeferenced with the same dataset of

control points used for the photogrammetric acquisitions. In

order to understand how the two different photogrammetric

projects were able to correctly represent the geometry of the

calibration field, a small sample area was chosen, and the two

photogrammetric clouds were compared with the one derived

from the laser scanner, using the C2C (Cloud to Cloud)

distances tool implemented in CloudCompare. The results of

these analyses are reported in Figure 12: the configuration A

resulted in a poor reconstruction of the geometry of the selected

portion (only the 4% of points present a deviation minor of

0.003 m from the LiDAR cloud) while the B configuration

reached good results (the 60% of points present a deviation

minor of 0.003 m from the LiDAR cloud). Is important to notice

that in the A configuration there are also several gaps in the

reconstruction of the geometry of the object.

Figure 11. Visual inspection of the Dense Cloud generated in

Photoscan. 360 configuration: automatic process with native

Exif (A), process with modified Exif (B).

Figure 12. C2C distances analysis performed in CloudCompare

with LiDAR data set as ground truth. Max distance set at 0.01 m.

360 configuration: automatic process with native Exif (A),

process with modified Exif (B).

4. DISCUSSION AND FUTURE PERSPECTIVES

The presented work aimed to evaluate the potentialities of the

tested 360 system for metric documentation purposes. Different

available configurations of the system were tested both during

the acquisition and the photogrammetric process. Starting from

a 3D calibration field carefully projected and using accurate and

precise control points, different strategies for the estimation of

camera interior orientation parameters were applied. One of the

most interesting output of this research can be traced in the

possibilities to estimate with a good level of confidence the

parameters of the six cameras performing a single acquisition

using the 360 rig configuration. The tables reported in the

Appendix demonstrate how the intervention of the operator can

be crucial in an automated photogrammetric workflow, and that,

especially in case of low-cost mass market sensors, the control

over these parameters is fundamental. Another important result

is the possibility to import the computed parameters of the six

cameras in the stitching software solution. The first test

performed underline that, also in these preliminary steps of

experimentation, an overall improvement of the spherical image

produced is appreciable also from a simple visual inspection.

Further and more robust analyses need to be performed in order

to quantify the impact of these parameters in the stitching

process. In the photogrammetric process the impact of the tested

practices was assessed with different analyses. Firstly, the

reprojection error of each tie points was extracted and analysed

for the two 360 configurations. Secondly, a comparison with a

more consolidated approach was achieved to evaluate the

quality of the processed point clouds in the reconstruction of the

object geometry.

Further test must be performed in this direction, a calibration on

a different 3D field is ongoing: the aim is to evaluate the

performances of the system in a wider environment, widening

also the camera-subject acquisition distance; to stress the



1118

operational features of the six cameras in different contexts.

Also, the acquisition in video mode need to be considered and

investigated and the use of other photogrammetric software

(with the relative calibration models included) must be

examined. Finally, the potentialities of the use of spherical

images in the photogrammetric process (i.e. spherical

photogrammetry) have risen the interest of several researchers:

Barazzetti et al., 2017a; Kossieris, Kourounioti, Agrafiotis, &

Georgopoulos, 2018.; Karol Kwiatek & Tokarczyk, 2015; K

Kwiatek & Tokarczyk, 2014; Ramos & Prieto, 2016 but a lot of

aspects need to be more in depth investigated and evaluate, due to

the constant technical evolution of sensors and software available

on the market.

REFERENCES

Balletti, C., Guerra, F., Tsioukas, V., & Vernier, P. (2014). Calibration of

action cameras for photogrammetric purposes. Sensors (Switzerland),

14(9), 17471–17490. https://doi.org/10.3390/s140917471

Barazzetti, L., Previtali, M., & Roncoroni, F. (2017a). 3D modelling with

the samsung gear 360, XLII(March), 1–3. https://doi.org/10.5194/isprs-

archives-XLII-2-W3-85-2017

Barazzetti, L., Previtali, M., & Roncoroni, F. (2017b). Fisheye lenses for

3D modeling: Evaluations and considerations. International Archives of

the Photogrammetry, Remote Sensing and Spatial Information Sciences -

ISPRS Archives, 42(2W3), 79–84. https://doi.org/10.5194/isprs-archives-

XLII-2-W3-79-2017

Fangi, G. (2015). Towards an easier orientation for spherical

photogrammetry. International Archives of the Photogrammetry, Remote

Sensing and Spatial Information Sciences - ISPRS Archives, 40(5W4),

279–283. https://doi.org/10.5194/isprsarchives-XL-5-W4-279-2015

Fangi, G., & Nardinocchi, C. (2013). Photogrammetric processing of

spherical panoramas. Photogrammetric Record, 28(143), 293–311.

https://doi.org/10.1111/phor.12031

Fraser, C. S. (1997). Digital camera self-calibration. ISPRS Journal of

Photogrammetry and Remote Sensing, 52(4), 149–159.

https://doi.org/10.1016/S0924-2716(97)00005-1

Fraser, C. S. (2013). Automatic Camera Calibration in Close Range

Photogrammetry. Photogrammetric Engineering & Remote Sensing,

79(4), 381–388. https://doi.org/10.14358/PERS.79.4.381

Gruen, A., & Beyer, H. A. (2001). System Calibration Through Self-

Calibration. In Calibration and Orientation of Cameras in Computer

Vision. (pp. 163–193). Springer, Berlin, Heidelberg.

https://doi.org/10.1007/978-3-662-04567-1_7

Holdener, D., Nebiker, S., & Blaser, S. (2017). Design and

Implementation of a Novel Portable 360 ° Stereo Camera System With

Low-Cost Action Cameras, XLII(November), 28–29.

Kossieris, S., Kourounioti, O., Agrafiotis, P., & Georgopoulos, A. (2017).

Developing a Low-Cost System for 3D Data Acquisition. ISPRS -

International Archives of the Photogrammetry, Remote Sensing and

Spatial Information Sciences, XLII-2/W8(November), 119–126.

https://doi.org/10.5194/isprs-archives-XLII-2-W8-119-2017

Kwiatek, K., & Tokarczyk, R. (2014). Photogrammetric applications of

immersive video cameras. https://doi.org/10.5194/isprsannals-II-5-211-

2014

Kwiatek, K., & Tokarczyk, R. (2015). Immersive photogrammetry in 3D

modelling. Geomatics and Environmental Engineering, 9(2), 51.

https://doi.org/10.7494/geom.2015.9.2.51

Läbe, T., & Förstner, W. (2004). Geometric stability of low-cost digital

consumer cameras. Proceedings of the 20th ISPRS Congress, Istanbul,

Turkey, 528–535.

Luhmann, T. (2004). A historical review on panorama photogrammetry.

International Archives of the Photogrammetry, Remote Sensing and

Spatial Information Sciences, 35(5), 8. https://doi.org/10.1.1.101.2588

Luhmann, T., Fraser, C., & Maas, H. G. (2016). Sensor modelling and

camera calibration for close-range photogrammetry. ISPRS Journal of

Photogrammetry and Remote Sensing, 115, 37–46.

https://doi.org/10.1016/j.isprsjprs.2015.10.006

Markiewicz, J. S., Łapi, S., Bienkowski, R., & Kaliszewska, A. (2017).

The example of using the xiaomi cameras in inventory of monumental

objects - first results, XLII(November), 28–29.

Perfetti, L., Polari, C., & Fassi, F. (2017). Fisheye photogrammetry: Tests

and methodologies for the survey of narrow spaces. International Archives

of the Photogrammetry, Remote Sensing and Spatial Information Sciences

- ISPRS Archives, 42(2W3), 573–580. https://doi.org/10.5194/isprs-

archives-XLII-2-W3-573-2017

Ramos, A. P., & Prieto, G. R. (2016). Only image based for the 3D metric

survey of gothic structures by using frame cameras and panoramic

cameras. International Archives of the Photogrammetry, Remote Sensing

and Spatial Information Sciences - ISPRS Archives, 41(July), 363–370.

https://doi.org/10.5194/isprsarchives-XLI-B5-363-2016

Remondino, F., & Fraser, C. (2006). Digital camera calibration methods:

considerations and comparisons. International Archives of

Photogrammetry Remote Sensing and Spatial Information Sciences, 36(5),

266–272.

Schneider, D., Schwalbe, E., & Maas, H. G. (2009). Validation of

geometric models for fisheye lenses. ISPRS Journal of Photogrammetry

and Remote Sensing, 64(3), 259–266.

https://doi.org/10.1016/j.isprsjprs.2009.01.001

Szeliski, R. (2006). Image Alignment and Stitching. Handbook of

Mathematical Models in Computer Vision.

Estimated interior orientation parameters of the six cameras separately considered and processed

A B C D E F

f(focale -mm) 3,049 3,042 3,040 3,040 3,044 3,035

f(focale -px) 1761,739 1757,888 1756,620 1756,499 1758,889 1753,511

cx (px) -59,928 -76,449 21,457 -1,514 48,974 50,579

cy (px) -23,881 -30,115 -44,285 -6,173 54,031 22,681

k1 (mm) 0,00533 0,00523 0,00496 0,00518 0,00518 0,00521

k2 (mm) 0,00018 0,00017 0,00021 0,00020 0,00019 0,00018

k3 (mm) -7,820E-06 -7,684E-06 -8,792E-06 -8,3076E-06 -8,068E-06 -7,652E-06

b1 (px) 0,0528 -0,4354 0,2083 -0,0041 -0,0546 -0,0416

b2 (px) -0,0820 -0,0213 0,0472 0,2613 0,0571 0,0862

p1 (mm) -2,640E-05 1,6157E-05 -8,781E-06 2,110E-05 1,993E-05 5,325E-06

p2 (mm) 3,470E-05 -1,490E-07 1,640E-06 -1,131E-06 -1,470E-07 1,667E-06

Table 2. Estimated interior orientation parameters of the six cameras separately considered and processed



1119

__________________________

* Corresponding author

APPENDIX

Figure 13. Camera parameters used in three different approach in AutoPano Giga.

360 Rig. Native Exif. Interior orientation parameters

f(focale -mm) 3,038

f(focale -px) 1755,593

cx (px) 0,381

cy (px) 1,070

k1 (mm) 0,00499

k2 (mm) 0,00021

k3 (mm) -8,878E-06

b1 (px) 1,0267

b2 (px) 0,4841

p1 (mm) 1,273E-05

p2 (mm) -9,230E-07

Table 3. Estimated interior orientation parameters of the six

cameras mounted on the rig and automatically processed with

the native Exif

360 Rig. Modified Exif. Estimated interior orientation parameters of the six cameras

A B C D E F

f(focale -mm) 3,050 3,038 3,040 3,041 3,045 3,035

f(focale -px) 1762,37711 1755,30797 1756,610 1756,98423 1759,15712 1753,58122

cx (px) -59,740 -76,220 21,893 -1,884 48,956 50,620

cy (px) -24,235 -30,213 -44,387 -6,310 54,120 22,333

k1 (mm) 0,00526 0,00620 0,00503 0,00516 0,00512 0,00521

k2 (mm) 0,00019 0,00001 0,00020 0,00020 0,00019 0,00018

k3 (mm) -0,00001 0,00000 -0,00001 -0,00001 -0,00001 -0,00001

b1 (px) -0,1357 -0,1955 0,1670 -0,2115 -0,0670 0,0277

b2 (px) -0,0682 -0,2481 -0,0361 0,2873 0,1400 -0,0095

p1 (mm) -4,89731E-07 1,33023E-05 -1,009E-05 2,237E-05 2,316E-05 1,340E-06

p2 (mm) 4,20493E-07 4,505E-07 1,868E-06 -9,839E-07 -2,416E-07 2,162E-06

Table 4. Estimated interior orientation parameters of the six cameras mounted on the rig and processed after the Exif modification



1120

PRELIMINARY EVALUATION OF A COMMERCIAL 360 MULTI-CAMERA … · PRELIMINARY EVALUATION OF A...

Documents

Transcript of PRELIMINARY EVALUATION OF A COMMERCIAL 360 MULTI-CAMERA … · PRELIMINARY EVALUATION OF A...